Dissociating viewpoint costs in mental rotation and object recognition